Reviewers are shared across halves; the reviewer analysis reads exposure, not a shared-reviewer split¶

reviewer-overlap.json (scripts/reviewer_overlap.py), metadata only, no outcome read. A change's reviewers are the pseudonymous accounts in its attention-set history other than its owner, change-level rather than the author of the example's own comment (the built examples do not keep it). Halves are the registered largest-first assignment; windows the corpus's own.

Exposure is the share of the sibling half's training examples that come from changes one of the dev example's reviewers reviewed (nearest-rank quantiles); 14 OpenStack and 36 Wikimedia dev examples sit in projects with no half and are left out, as the placebo corpus leaves them out. Attention sets can hold CI or bot accounts, and no raw account carries a service tag, so they cannot be filtered; one Wikimedia account attends 1,978 changes and owns two. The two-reviewer row is the check against that, since no single account can make an example count there. Almost every dev example shares some reviewer with its sibling half, so the shared-reviewer split ROADMAP Plan H planned would leave one side nearly empty, and a foreign organization shares none: the sibling-over-foreign gain the house-style claim rests on is the place a reviewer effect could hide. Exposure varies over two orders of magnitude, so the analysis to register is a moderation of the sibling-over-foreign contrast by exposure (does the sibling's advantage grow with how much of its training its own reviewers wrote), as a secondary before the seal, change-clustered as the primary contrast is.